Research On Network Traffic Classification Based On Machine Learning

With the rapid development of Internet,it has become a vital part of people's production and life.As an important part of network management system,network traffic identification is the basic technology for various net,work applications and network management.As the size of the network grows,the variety of network applications continues to grow,and the network environment becomes increasingly complex.The network traffic identification method based on flow statistics can be combined with Machine Learning(ML)technology to effectively solve the defects of traditional methods.However,existing ML-based traffic identification methods still have shortcomings.In response to these problems,this thesis studies the network traffic identification based on deep learning technology.The main research results are as follows:1.An incremental learning network traffic classification model based on Convolutional Neural Network(CNN)is studied and proposed.On the one hand,the method compares the traffic data into images and directly uses the flow binary data as the model input,which solves the problem of difficult feature selection in the existing methods;On the other hand,we present a complete incremental learning algorithm by introducing the proficiency mechanism.It solves the potential problem of catastrophic forgetting and stability-plasticity trade-offs in incremental learning.2.An online learning method based on combined model is studied and proposed.This model can identify traffic in early stage.Based on the incremental learning model,a semi-supervised flow level model is implemented based on LSTM and CFSFDP clustering.The incremental learning model is used as a packet level model.On the basis of the semi-supervised model,a filtering mechanism is added as a flow level model.The filtered flow level result is used as the label to train the packet level model in an online manner.Finally,based on the combined model,an early stage identification method with online learning ability is realized.The experimental results show that the online learning network traffic identification model proposed in this thesis not only has the ability to automatically adapt to changes in the network environment,but also provides semi-supervised features and early identification of traffic.This method improves the practicability of the network traffic recognition model based on deep learning and reduces the cost of the model application.